REALITY and XTLS Vision Explained: TLS Handshake Overhead and Encryption Efficiency

Learn how TLS handshakes work, why REALITY removes certificate management for proxy servers, and how Vision flow control can reduce redundant encryption—and what that means for latency, CPU use, and mobile battery life.

At a glance

For readers who can already import VLESS subscriptions and want to understand node fields and speed-test results. REALITY changes how the proxy entry authenticates clients; Vision optimizes data forwarding under specific conditions. REALITY does not necessarily eliminate a network round trip, and Vision does not turn HTTPS into plaintext. Understanding the distinction helps pinpoint whether an effect comes from connection setup, sustained transfers, or device power use.

The TLS 1.3 handshake: What happens before the first byte

TCP connectionClient helloServer handshakeAuthenticationApplication data

With standard HTTPS over TLS 1.3, the client uses ClientHello to offer supported protocol parameters and key shares. The server selects the parameters, returns its own key share, and sends a certificate and signature in subsequent handshake messages. The client verifies the certificate chain, hostname, and handshake signature. Once both sides confirm the handshake records, application data can be sent as usual. A certificate proves identity; it does not encrypt the entire webpage.

When accessing HTTPS through a proxy, there may be two separate secure sessions: one between the client core and the proxy entry, and another between the application and the destination website. The browser still verifies the website's certificate on the latter connection. A page's time to first byte also includes the proxy-to-destination connection, server processing, and possibly DNS lookups, so it is not a direct measure of proxy handshake time.

1 RTT
Standard TCP connection setup
1 RTT
Standard full TLS 1.3 handshake
160 ms
Theoretical example: two round trips at 80 ms RTT

How to measure: test new and reused connections separately

160 ms is an idealized estimate for a single connection segment, not a measured time to first render. It excludes DNS, packet loss and retransmissions, server processing, and the connection to the destination site. If only new connections are slow, check round-trip latency and retransmissions first, then investigate the handshake. Don't infer connection setup overhead from the average speed of a single large download.

REALITY without certificates: no certificate management for the proxy entry

A standard TLS entry typically requires a certificate for its server name, private-key protection, and renewals. REALITY does not require proxy operators to obtain and renew a public CA certificate specifically for the entry. The client authenticates the proxy using REALITY public-key material obtained in advance; the server keeps the corresponding private key. What changes is the source of trust—not the need for authentication. This is not a case of disabling certificate checks and connecting anyway.

XTLS Vision flow control: less redundant processing

In a traditional nested path, the application's TLS ciphertext is encapsulated again by the proxy's security layer. After the proxy entry removes the outer layer, the data forwarded to the destination is still encrypted at the application layer. These two layers have different endpoints; double encryption does not mean the application verifies two certificates for the same connection.

Vision is a VLESS flow-control option in Xray, with the value xtls-rprx-vision. When the implementation can recognize inner TLS and the conditions for switching are met, subsequent data can avoid redundant encryption and decryption in the outer layer, while the inner TLS session remains between the application and destination website. The fast path does not automatically apply during connection setup or to payloads that don't meet the requirements. In particular, plaintext traffic must not be treated as if it were protected by inner TLS.

Combining REALITY and Vision

Proxy protocol
VLESS
Transport
RAW/TCP
Security layer
REALITY
Flow
xtls-rprx-vision

These fields describe the protocol, transport, security layer, and flow control separately. Check each one after importing the configuration.

Standard VLESS over TLS

Proxy protocol
VLESS
Transport
RAW/TCP
Security layer
TLS
Flow
Not set

This card shows a comparison without Vision enabled; it does not mean TLS and Vision are mutually exclusive.

Speed, CPU use, and battery life: measure each separately

Reducing an encryption and decryption layer can lower per-byte processing costs in the core, but throughput is likely to change noticeably only when processing is near the device's bottleneck. If link bandwidth, proxy egress limits, or the destination's response time is the constraint, CPU use may drop while download speed stays almost the same. On mobile devices, account for screen use, wireless signal strength, and how long the radio stays active after a transfer ends. Lower core CPU use does not necessarily mean lower overall power consumption.

  1. Time to first byte: Use a controlled test environment with the same device, server, and HTTPS resource. Run 10 trials with new connections, then record a separate set with connection reuse. Report the median and maximum, and note any network changes or retries during testing.
  2. Throughput and CPU: Use the same fixed 100 MiB resource, alternating between the two test groups. Record completion time and the core process's cumulative CPU time. Compare CPU time per MiB rather than relying on a single Task Manager percentage reading.
  3. Mobile battery use: Keep screen state, network type, and signal conditions consistent. Compare runs using the same data volume or the same duration, and state which method you used. Battery percentage changes over a few dozen seconds cannot reliably distinguish two flow-control options and should not be used to claim a fixed percentage saving.

How to check the core and node settings after importing

Why does the core report an unsupported configuration after importing a subscription?

Being able to save a node only means the client read the entry. On desktop, check the core-type options under v2rayN's Settings → Parameters, then review the actual startup log. Menu names may vary by version. On Android, check both the app and core: v2rayNG uses Xray, while v2flyNG uses V2Fly. Don't assume a REALITY and Vision configuration intended for Xray can be copied across cores.

Handshake failed. Can I clear the Flow field first?

Open the node's edit screen and compare its address, remote port, security layer, server name, and public-key material with the original subscription. Then check whether the first failure occurs during connection setup or the handshake. Don't change Flow on only one side to test the server's requirements; use a comparison node explicitly identified by the provider if you need one.

The node shows port 443. Can I use it as my local proxy port?

Not directly. The node's 443 is an example of a remote port. When configuring a proxy manually in an application, use the local address and port the client is actually listening on. Check the local listener in the client's settings, and distinguish the HTTP and SOCKS endpoints. Don't enter the remote address in the local proxy fields.

Why is the webpage still slow when downloads max out?

Large downloads mainly measure sustained transfer speed. Webpages also depend on DNS, initial connection setup, and multiple small requests. In the same browser, compare a page's first load with an immediate reload, then use the developer tools' Network panel to see which requests are spending time waiting. If only one domain is slow, investigate its requests and routing first; don't change flow control based on download tests alone.

Download the client